TCE, Georgia Tech Collaborate on Agentic AI
Automated laboratories have dramatically accelerated material synthesis, allowing scientists to produce libraries of hundreds or thousands of new materials in one go. However, characterizing each of those materials still requires scientists to individually conduct measurements on multiple devices, which can take months.
“New materials underpin technologies ranging from energy generation and storage to electronics, manufacturing, quantum technologies, and healthcare,” said Mahshid Ahmadi, an associate professor in the Department of Materials Science and Engineering (MSE). “Yet discovering and optimizing those materials remains slow and expensive.”
This summer, Ahmadi received a $60,000 Scialog: Automating Chemical Laboratories research award to bridge that gap. The initiative, funded by the Research Corporation for Science Advancement (RCSA), the Arnold and Mabel Beckman Foundation, and the Frederick Gardner Cottrell Foundation, was created to accelerate advances in automated laboratory technologies.
Ahmadi and Georgia Tech Assistant Professor Vida Jamali, who received her own $60,000 award, proposed creating an agentic AI that will help researchers sift through experimental data and make scientifically informed decisions about which materials to investigate further. Their team was one of just seven to receive Scialog funding this year.
“I am very excited and honored that our project was selected,” Ahmadi said. “I see this award as part of a much larger transformation happening in experimental science.”
In addition to Ahmadi and Jamali’s combined expertise, their collaboration will also benefit from ATHENA (Advanced Testbed for High-throughput Experimentation in Nano- and Atomic Science), the University of Tennessee’s new National Science Foundation-funded hub for AI-powered laboratories. Ahmadi is one of the lead investigators bringing ATHENA to life.
“ATHENA is building a broader framework for AI-enabled, distributed experimentation and accelerated materials characterization,” Ahmadi said. “This Scialog project addresses one of the central scientific challenges within that vision while developing Tennessee’s leadership in autonomous science.”
Investigating Smarter, Not Harder
The RCSA founded Scialog (a portmanteau of “science” and “dialog”) in 2010 to stimulate interdisciplinary advancement in scientific issues of global importance.
Ahmadi was one of just 45 early-career scientists invited to the RCSA’s 2026 Scialog meeting in April of this year. Working with colleagues they had not previously collaborated with, the participating scientists developed projects that would close the gap between the rapid pace of automated instrumentation and the lagging rate of experimental work.
Researchers often have to choose between conducting relatively simple measurements across an entire materials library—potentially missing important traits only visible at high resolution—or performing expensive, sophisticated characterization on a small subset of samples that might not ultimately be useful.
Ahmadi and Jamali’s plan will create an agentic AI that scans through fast, relatively inexpensive measurements and identifies samples that merit more sophisticated characterization. High-resolution characterization methods will thus be reserved for the most promising materials or those that produced conflicting early results.
“Our key innovation is that we are not simply using AI to analyze data after an experiment,” Ahmadi explained. “We are developing agentic AI that participates in deciding how characterization itself should proceed.”
Building the Agentic AI
To construct and test their agentic AI framework, Ahmadi and Jamali need to generate training data and give the AI a chance to make scientifically relevant recommendations.
Ahmadi’s lab at UT will create a library of halide perovskites, which are highly tunable materials commonly used in LED lights and solar panels. After conducting optical spectroscopic and structural measurements of the library, Ahmadi’s lab members will develop specialized AI agents that can interpret the data, generate hypotheses, and design relevant protocols to further investigate the most promising perovskites.
“One of the most exciting aspects for our students is that this project sits at the intersection of materials science, chemistry, AI, robotics, and data science,” Ahmadi said. “They will learn not only how to perform sophisticated experiments, but also how to think about experiments as part of an intelligent, adaptive system.”
Meanwhile, Jamali’s team at Georgia Tech will develop approaches to align X-ray diffraction, optical imaging, and spectroscopy data streams to facilitate AI analysis. Ahmadi and Jamali will share data descriptors and metadata over ATHENA, testing the AI agent’s ability to intelligently request increasingly sophisticated measurements only when they are scientifically necessary. When the AI agent flags a material for additional study, Jamali’s lab will conduct high-resolution transmission electron microscopy measurements to validate the recommendation.
Using ATHENA in this project will also give students at both universities vital experience working in autonomous science ecosystems, which Ahmadi expects will only become more prevalent as AI agents become more sophisticated.
“This project could substantially shorten the time between a scientific question and the discovery of a useful material or chemical system,” Ahmadi said. “If autonomous laboratories can not only synthesize materials faster but also measure them intelligently in a way that fosters cross-institutional collaboration, we can reduce unnecessary experiments, make better use of expensive scientific instrumentation, and accelerate the cycle of discovery.”
Contact
Izzie Gall ([email protected])